chank@cb.ecn.purdue.edu (King Chan) (10/05/90)
Greetings, I would like propose two neural network questions. Perhaps someone can give me some insight as to their solution. I. Input Representation for Variable Length Data It appears that most nn inputs are of fixed length (i.e the number of input neurons are static). However, there are cases where this is not applicable. For instance, a 2-D connection table representation of chemical structure will vary in size depending on the molecule. Is there a way to get around this ? II. Functional Relationship Training with Missing Data Has there been any work on NN training and recall when all the input and/or output node information is unavailable. It appears to me that a value of 0 for an unknown attribute value is insufficient. Any Comments ? I hope these questions merit some discussion. Thank you very much. chank@cn.ecn.purdue